🤖 Multi-Skill Reinforcement Learning for the Unitree G1 Humanoid in MuJoCo

In this implementation, the Unitree G1 humanoid learns multiple behaviours through PPO-based reinforcement learning using MuJoCo Playground, Brax PPO, and JAX/CUDA acceleration.

The robot was trained to perform:
• Stable humanoid walking, slow and fast
• Vertical jumping

Training was performed across more than 230 million environment steps using massively parallel GPU simulation on RTX 3000 Ada and RTX 4090 GPUs.

This project explores reward engineering, humanoid locomotion, sim-to-real RL pipelines, and scalable GPU-based policy training for high-DoF robots.

Tech Stack:
• MuJoCo Playground
• Brax PPO
• JAX/CUDA
• PPO Reinforcement Learning
• Unitree G1 Humanoid

#Robotics #ReinforcementLearning #HumanoidRobot #MuJoCo #PPO #MachineLearning #AI #Unitree #JAX #Brax #RobotLearning #Sim2Real

Видео 🤖 Multi-Skill Reinforcement Learning for the Unitree G1 Humanoid in MuJoCo канала Learning Plus To Date (A. Zaraki)